Hybrid Data-Mechanistic Approach to Sensor Fault Detection for Direct-Drive Electric Vehicle Motors

Sensor fault diagnosis for direct-drive electric vehicle in-wheel motor systems faces significant challenges under complex operating conditions. Specifically, signal disturbances such as noise interference, model uncertainty, and environmental factors often cause residual distortion, masking fault features and increasing false alarm rates. To address these challenges, a fault diagnosis method based on data-driven subspace identification and residual anti-interference filtering is proposed. Firstly, a state space model of the permanent magnet brushless DC motor is established, and a subspace identification algorithm is combined to construct a discrete system model considering disturbances. An adaptive residual filter is designed to suppress the influence of disturbances on residual signals, and predictive time-domain rolling optimization of residual estimation is introduced. Furthermore, the maximum likelihood ratio is used to construct adaptive thresholds to enhance fault sensitivity and robustness. The bench and real vehicle experiments show that this method can effectively extract fault features under noise and uncertainty interference, significantly improve the fault localization accuracy of current and speed sensors, reduce false alarm rates, and provide technical support for the reliable operation of direct-drive motor systems.